0
$\begingroup$

I want to test whether an individual's grouping (categorical of which there are 4 separate groups) together with their response to another question (likert type, so assumed ordinal) affects their preference for a range of strategies (which they individually ranked from 1-6)

Would either ordinal regression or multivariate analysis be suitable to test this? My data is not normally distributed, so would need to use non-parametric tests.

$\endgroup$
5
  • $\begingroup$ Sounds like two have two categorical variables, and you want to test association. Anything wrong with chi-square test? $\endgroup$
    – SmallChess
    Apr 5, 2017 at 11:55
  • $\begingroup$ In your question, you don't have enough variables for multivariate analysis. But you can just use the ordinal as response for a ordinal regression model. $\endgroup$
    – SmallChess
    Apr 5, 2017 at 11:57
  • $\begingroup$ Thank you - yes I realised after I posted that for a multivariate analysis I would need two DVs, and I only have one, so you would say ordinal regression is best to use for influence of two IVs (together) on one DV? $\endgroup$
    – Dragonfly
    Apr 5, 2017 at 11:58
  • $\begingroup$ From the OP I understand there to be 6 dependent variables: each strategy that receives a rank. $\endgroup$
    – rolando2
    Apr 5, 2017 at 13:08
  • $\begingroup$ @rolando2 Maybe. That could also be a single ordinal variable, with values from 1-6. $\endgroup$
    – SmallChess
    Apr 5, 2017 at 13:09

1 Answer 1

3
$\begingroup$

As noted in the comments, multivariate analysis isn't possible here.

When the dependent variable is ordinal (as yours is) then ordinal logistic regression is a good starting point, at least, but it makes an assumption of proportional odds. If that assumption is violated, then the most common choice is multinomial logistic regression, but there are some other choices too, such as the partial proportional odds model.

You might also try a tree, or one of the methods based on trees.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.